Ultra-fast silicon detectors (UFSD) · 2018. 9. 28. · cision Proton Spectrometer (CT-PPS), residing in Roman-pots about 200 m from the interaction region. In both cases, the UFSD

Nuclear Instruments and Methods in Physics Research A 831 (2016) 18–23

Contents lists available at ScienceDirect

Nuclear Instruments and Methods inPhysics Research A

http://d0168-90

n CorrE-m

journal homepage: www.elsevier.com/locate/nima

Ultra-fast silicon detectors (UFSD)

H.F.-W. Sadrozinski a,n, A. Anker a, J. Chen a, V. Fadeyev a, P. Freeman a, Z. Galloway a,B. Gruey a, H. Grabas a, C. John a, Z. Liang a, R. Losakul a, S.N. Mak a, C.W. Ng a, A. Seiden a,N. Woods a, A. Zatserklyaniy a, B. Baldassarri b, N. Cartiglia b, F. Cenna b, M. Ferrero b,G. Pellegrini c, S. Hidalgo c, M. Baselga c, M. Carulla c, P. Fernandez-Martinez c, D. Flores c,A. Merlos c, D. Quirion c, M. Mikuž d, G. Kramberger d, V. Cindro d, I. Mandić d, M. Zavrtanik d

a SCIPP, Univ. of California Santa Cruz, CA 95064, USAb Univ. of Torino and INFN, Torino, Italyc Centro Nacional de Microelectrónica (CNM-CSIC), Barcelona, Spaind IJS Ljubljana, Slovenia

a r t i c l e i n f o

Article history:Received 25 November 2015Received in revised form22 March 2016Accepted 29 March 2016Available online 31 March 2016

Keywords:Fast silicon sensorsCharge multiplicationThin tracking sensorsSilicon stripPixel detectors

x.doi.org/10.1016/j.nima.2016.03.09302/& 2016 Elsevier B.V. All rights reserved.

esponding author.ail address: [email protected] (H.-W. Sadrozi

a b s t r a c t

We report on measurements on Ultra-Fast Silicon Detectors (UFSD) which are based on Low-Gain Ava-lanche Detectors (LGAD). They are n-on-p sensors with internal charge multiplication due to the pre-sence of a thin, low-resistivity diffusion layer below the junction, obtained with a highly doped implant.We have performed several beam tests with LGAD of different gain and report the measured timingresolution, comparing it with laser injection and simulations. For the 300 μm thick LGAD, the timingresolution measured at test beams is 120 ps while it is 57 ps for IR laser, in agreement with simulationsusing Weightfield2. For the development of thin sensors and their readout electronics, we focused on theunderstanding of the pulse shapes and point out the pivotal role the sensor capacitance plays.

& 2016 Elsevier B.V. All rights reserved.

1. Introduction

We propose an ultra-fast silicon detector that would establish anew paradigm for space-time particle tracking [1]. Presently,precise tracking devices determine time quite poorly while goodtiming devices are too large for accurate position measurement.We plan to develop a single device that ultimately will measurewith high precision concurrently the space (�10 μm) and time(�10 ps) coordinates of a particle.

First applications of UFSD are envisioned in LHC upgrades, incases where the excellent time resolution coupled with goodspatial resolution helps to reduce drastically pile-up effects due tothe large number of individual interaction vertices. While ATLAS isproposing UFSD as one of the technical options for the HighGranularity Timing Detector (HGTD) located in front of the for-ward calorimeter (FCAL), CMS-TOTEM are considering UFSD to bethe timing detectors for the high momentum - high rapidity Pre-cision Proton Spectrometer (CT-PPS), residing in Roman-potsabout 200 m from the interaction region. In both cases, the UFSDwould be of moderate segmentation (a few mm2) with challenging

nski).

radiation requirements (few times 1015 neq/cm2), requiring a timeresolution of 30 ps, which could be achieved by stacking up inseries up to four sensors.

UFSD are thin pixelated n-on-p silicon sensors based on theLGAD design [2,3] developed by CNM Barcelona. The LGADs ex-hibit moderate internal gain (�10x) due to a highly dopedpþregion just below the n-type implants. Based on the progressmade through 7 fabrication cycles, the performance of LGAD havebeen established in several beam tests and with laser laboratorymeasurements. The sensors tested were routinely operated forlong time periods at an operating bias voltage close to 1000 V for300 mm thickness (500 V for 50 mm) and various internal gains of3–20.

Since present experience with LGAD is limited to sensors with300 μm thickness [4], a reliable tool is needed to extrapolate theirperformance to the planned thickness of 50 μm. This is done withthe simulation program Weightfield2 (WF2) [5] that has been de-veloped specifically for the simulation of the charge collection insemiconductors. In the following, we compare the pulse shapes ofthick and thin LGAD to elucidate the advantage of thin sensors,including those due to trapping effects after irradiation. This isfollowed by an introduction to precision timing in silicon detectorsand a prediction of the expected timing resolution as a function ofLGAD thickness and internal gain. The predictions will be

H.-W. Sadrozinski et al. / Nuclear Instruments and Methods in Physics Research A 831 (2016) 18–23 19

confronted with results from several beam tests and laboratorylaser measurements. Finally we present pulse shapes on thinLGADs and the pivotal role the sensor capacitance plays in thetiming resolution of UFSD.

2. LGAD pulse shapes

The Weightfield2 program [5] simulates the electrostatic fieldsand the charge collection in LGAD, including the effect of the in-ternal gain. The current output of the sensor can then be con-voluted with the response of the front-end electronics generatinga voltage signal that can be used to evaluate the timing capabilitiesof a detector. Fig. 1.a shows the output current for a minimumionizing particle (MIP) traversing a 50 μm thick LGAD with gain 10biased at large over-depletion, showing the separate contributionsfrom the drift of both the initial and gain electrons and holes,respectively. For thicker LGAD, the current pulse has the sameshape as that shown in the picture, with the only difference thatthe pulse duration is scaled by the thickness, i.e. the 1 ns collectiontime for the 50 mm thick LGAD becomes 9 ns for 300 mm thickness.In Fig. 1.b the voltage signals from a broad-band amplifier (BB) areshown for LGADs of different thickness, indicating that for con-stant gain the maximum pulse height is independent of the LGADthickness, and that the shorter rise time favors the thin sensor fortiming application.

The change of the LGAD pulse shape due to trapping after ir-radiation can be studied with WF2, of which version 3.5 in-corporates trapping [6]. Since the characteristic trapping time isabout 0.5 ns (corresponding to a trapping length of �50 mm), oncomparing the signals from thin and thick detectors shown inFig. 1.b one would expect that the longer pulses of thick detectorwill be effected much more by trapping than the short ones fromthin LGAD. This is illustrated in Fig. 2 where the BB pulses forLGAD with gain 10 and thickness a) 300 mm and b) 50 mm, re-spectively, (note the different time scale) are shown for differentneutron fluences. For 300 mm LGAD (Fig. 2.a), the large loss of gainholes changes the pulse shape drastically and reduces the ob-served gain (defined as the ratio of pulse areas of LGAD over thatof no-gain diodes) by a large amount. The effect of trapping onthin sensors is much less drastic as shown in Fig. 2.b: the pulseshape and the rising edge are preserved (which is good for timing)and the gain loss is limited.

For timing application, the pulse amplitude is more importantthan the pulse area. The variation of signal amplitude as a functionof neutron fluence is shown in Fig. 3 for 300 and 50 mm thickLGADs: up to a fluence of 4�1015, the pulse height loss due to

Fig. 1. Pulse shapes of LGAD simulated with WF2 version 3.5: a) detector current foramplifier (BB) with 50 Ω input for LGADs with gain of 10 and thickness 50, 150, 300 m [

trapping for a 50 mm thick LGAD is less than 50% of its pre-radvalue.

The mechanisms underlying the radiation effects in LGADs areunder intensive investigation within RD50 [7]. Up to now, data areavailable for 300 mm thick LGAD, and the data are interpreted interms of a decrease in the gain in addition to the signal decreasecaused by trapping at fluences beyond 1014 neq/cm2 [8]. This hasbeen identified with an initial acceptor removal, depending onboth the boron doping concentration and the interstitial defectscreated during irradiation [9]. The acceptor removal appears tolevel off at higher fluences so that a gain of about 3.5 is observed ata fluence of 2*1015 neq/cm2, for which we project a timing re-solution of about 60 ps, using Figs. 4 and 7 and assuming that thetiming resolution scales with dV/dt. We are fabricating thin sen-sors with a variety of gain values and bulk resistivities for irra-diations to verify the acceptor removal model. In addition, we areworking on replacing the boron in the multiplication layer bygallium, which has been shown to be more radiation resistant.

3. Simulation of the UFSD timing resolution

We have used WF2 to simulate LGAD parameters which drivethe timing resolution: internal gain, capacitance and thickness.The time resolution st is given by contributions from time walk,jitter and TDC binning:

σ = + +⎛⎝⎜⎜

⎡⎣⎢

⎤⎦⎥

⎞⎠⎟⎟

⎛⎝⎜

⎞⎠⎟

⎛⎝⎜

⎞⎠⎟dV dt

NdV dt

TDCV/ / 12

2t

th

RMS

2 2bin

2

with Vth the signal threshold, dV/dt the signal slope or slew-rate, Nthe noise, and TDCbin the size of a TDC bin, indicating the centralrole of the slew-rate of the signal dV/dt [10]. This means that weneed both large and fast signals. We are still quantifying thecontributions to the time resolution due to the non-uniformcharge deposition within the sensor caused by local Landau fluc-tuation (in addition to the standard time-walk contribution), andwill report on this issue soon in a separate paper. Using WF2, wecan show that the time resolution improves with larger gain aswell as with thin detectors (Fig. 4), since both increase the slew-rate. An additional advantage is expected from sensors with re-duced capacitance, i.e. small area, as they permit larger slew-ratefor a fixed input impedance of the amplifier (see Section 5 below).

4. Timing resolution measurements

We measured the time resolution of 300 mm thick LGAD pads

a MIP traversing a 50 mm thick LGAD; b) voltage output from a x100 broad-band5].

Fig. 2. WF2 simulation of BB pulse shapes of MIP signals due to trapping for different neutron fluences (in units of neq/cm2) for LGAD of gain 10 with two thickness’: a)300 mm, b) 50 mm. Note the different time scales.

Fig. 3. WF2 simulation of the BB pulse height of MIP signals as function of neutronfluence for LGAD of gain 10 with 50 mm and 300 mm thickness when only trappingis considered.

Fig. 4. WF2 simulations of the slew-rate dV/dt as measured by a 50 Ω Broadbandamplifier as a function of sensor thickness and various gain values. They indicatethe good time resolution achievable with thin LGAD with gain. At 50 mm thickness,a gain of 10 results in a three-fold improvement in the time resolution whencompared to a no-gain sensor.

Fig. 5. Timing resolution and time walk of the mean at constant 10 mV thresholdfor a 300 mm thick LGAD with gain 10 in the Nov. 2014 beam test. The vertical linesindicate the range of a MIP. A running average of 1 ns is used to filter the data [11].

H.-W. Sadrozinski et al. / Nuclear Instruments and Methods in Physics Research A 831 (2016) 18–2320

with internal gains between 10 to 20 in the CERN H6 170 GeV pionbeam using sensors with different capacitances (4 pF and 12 pF)[11]. With a view on the upcoming design of the electronicsreadout, we used several analysis algorithms to optimize the time

resolution: (i) a constant low threshold, (ii) the time of the pulsemaximum, (iii) an extrapolation of the slope to the base line, and(iv) a constant fraction discriminator (CFD). As an example, Fig. 5shows the timing resolution and time walk at constant 10 mVthreshold as a function of pulse height for a 300 mm thick LGAD. Inthe region of single MIPs with pulse height between 40 and80 mV, the timing resolution is between 150 ps and 200 ps, andthe time walk is substantial at about 400 ps.

Fig. 6 shows the time resolution for LGADs with different ca-pacitances (12 pF the sensor of Fig. 5 and 4 pF) as a function of CFDthreshold for different filter (running average) times of the pulse.The CFD approach is the timing method we found preferable sinceit eliminates to a great extent the time walk error [11]. As pre-dicted by simulations, reducing the capacitance from 12 pF to 4 pFimproves the time resolution by 25%, going from 160 ps to 120 psfor CFD threshold set at 5–15% and applying a low pass filter set atabout 500 MHz.

Our present understanding of the timing resolution for 300 mmthick LGAD is shown in Fig. 7. Improved resolution is seen for laservs. beam test data since the laser is not subject to time walk andLandau fluctuations. Another improvement measured and prop-erly predicted is when the LGAD capacitance is reduced. The onlymeasurement not agreeing with the WF2 simulations is the lowestlaser measurement at 300 mm. The fact that it is lower than theWF2 prediction is traced to an improved noise behavior of themeasurement not captured in the simulations [10]. The goodagreement of the measured time resolution from both lasermeasurements (only time jitter) and beam tests (time jitter, time

Fig. 6. CFD Time resolution of LGADs with different capacitances: a) 12 pF (left) [11] and b) 4 pF.

Fig. 7. Time resolution for LGADs with gain of 10 as a function of sensor thickness, combining both test beam and laser measurements (at 300 mm) – closed symbols-withWF2 simulations – open symbols-.

H.-W. Sadrozinski et al. / Nuclear Instruments and Methods in Physics Research A 831 (2016) 18–23 21

walk and Landau fluctuations) with the WF2 simulation justifiesthe extrapolation of the expected time resolution to thinner sen-sors (Fig. 7). For a 50 mm thick LGAD with gain of 10 we expect atime resolution of 30 ps.

5. Thin LGAD

Thin LGAD were produced on 100 Ω-cm epitaxial p-type waferswith different pad areas, and used to investigate the effect of thecapacitance on the output pulses. Fig. 8 shows the results of

Fig. 8. Measurement results on 50 mm thick epitaxial LGAD: a) C–V measurement showinFZ and epi LGAD indicating lower doping concentration in the multiplication layer for thno-gain diode yielding a gain of 3.5 for the LGAD.

measurements on two 50 mm thick pads: the 1/C2 curve (Fig. 8.a)indicates a depletion voltage of 170 V and the capacitances to be2.6 pF for the small diode SD2 of area 1 mm�1 mm and 35 pF forthe large diode BD3 of area 4 mm�4 mm. It also shows the“voltage lag” at low voltages typical for LGAD accounting for thedepletion of the gain layer. When the C–V data are used to extractthe doping concentration (Fig. 8.b), the epi LGADs show a lowervalue of the gain layer doping with respect to what has been seenin previous float zone LGAD with gain of 15 and 7, respectively [4].The measured gain of 3.5 using IR laser shown in Fig. 8.c is con-sistent with the doping profile and with the relatively small

g a relatively small “voltage lag” at low bias; b) doping profile extracted from C-V fore epi LGAD; c) comparison of charge collection in IR laser injection on epi LGAD and

Fig. 9. Response to front α particle injection of 50 mm thick epitaxial LGADs: a) pulse shapes including WF2 simulations for different capacitances and gain of 3.5; b) slew-rate as a function of capacitance for LGAD with gain 10 and 15.

Fig. 11. Time jitter vs. LGAD capacitance for two noise values at the amplifier input:N¼18 mV (presently measured) and N¼ 10 mV (goal).

H.-W. Sadrozinski et al. / Nuclear Instruments and Methods in Physics Research A 831 (2016) 18–2322

voltage lag of Fig. 8.a. The data show a bias voltage range of 500 V,very large for the thin sensors.

Our measurement is performed using a broad-band amplifier offixed 50 Ω input impedance; for our analysis we need to properlytake into account the effect of the sensor capacitance C. A firstimportant effect is that the capacitance of the LGADs has a stronginfluence on the pulse shapes: see Fig. 9.a for pulse shapes takenwith α particles injected from the front of the sensors togetherwith the corresponding simulated WF2 pulses. The simulated WF2data shown are displayed on a vertical scale which has beenproperly adjusted to take into account the gain of the amplifierand the fraction of the α energy absorbed in the sensitive part ofthe LGAD, about 50%, as determined from the collected charge.Compared to a LGAD with C¼35 pF, the small LGAD with C¼2.6 pFexhibits a 3-fold increase in amplitude and a 5-fold increase in theslew rate dV/dt as seen in Fig. 9.b.

A second effect in the use of a broad-band amplifier is that thenoise N is independent of the LGAD capacitance. We changed thecapacitance between 2.6 and 223 pF by ramping up the bias vol-tage, and measured the RMS noise on random triggers using dif-ferent low-pass bandwidth (BW) limits on the digital scope: asshown in Fig. 10 the noise RMS does not change over this largerange of capacitances. At the highest bias beyond 400 V, an in-crease of noise due to the leakage current is observed. For all ca-pacitances, the bandwidth dependence of the noise varies like(BW)0.4. We find N(1 GHz)¼18 mV at the amplifier input.

The fact that the noise is independent of the capacitance allowsus to calculate the time jitter, i.e. part of the timing resolution due

Fig. 10. Noise RMS of the SD2 epi LGAD for different Bandwidth limits when thecapacitance is varied by changing the bias voltage.

to the noise, for different LGAD capacitances, by dividing the noiseby the slew-rate (Fig. 9.b):

σ =⎛⎝⎜

⎞⎠⎟

NdV dt/Jitter

The time jitter vs. LGAD capacitance is shown in Fig. 11 for twonoise values at the amplifier input: N¼18 mV (presently measured)and N¼10 mV (goal). A time jitter of 10 ps seems achievable forsmall capacitances, while the jitter for LGAD with C¼10 pF canreach below 20 ps. As shown in Fig. 7, the time jitter constitutesthe largest part of the time resolution. To set the scale, a2 mm�2 mm sensor, 50 mm thick, has a capacitance of 8 pF.

6. Conclusions

We measured the timing resolution of 300 mm thick Low-GainAvalanche Diodes and found 120 ps in a beam test and 65 ps withan IR laser. Both numbers are in agreement with Weightfield2 si-mulations. The same simulation program predicts a timing re-solution of 30 ps for 50 mm thick LGAD of 2 pF capacitance.

Of the different methods used to determine the time stamp of apulse, the constant-fraction discriminator shows the bestperformance.

We use 50 mm thick epitaxial LGAD with low gain to investigatethe effects the sensor capacitance has on the pulse height and the

H.-W. Sadrozinski et al. / Nuclear Instruments and Methods in Physics Research A 831 (2016) 18–23 23

slew-rate dV/dt, which is the main parameter determining thetiming resolution of UFSD. When a broad-band amplifier is used,an increase of the capacitance from 2.6 pf to 35 pF decreases thepulse height by a factor 3 and the slew-rate by a factor 5.

With a broad-band readout, the LGAD noise is independent ofsensor capacitance, and varies like BW0.4 as a function of the band-width of a low-pass filter. Work to reduce the noise by a factor 2xbeyond the presently achieved level is ongoing.

Acknowledgments

We thank HSTD10 Organizing Committee for their hospitalityand the high scientific standards of the Symposium.

We acknowledge the expert contributions of the SCIPP tech-nical staff. Part of this work has been performed within the fra-mework of the CERN RD50 Collaboration.

The work was supported by the United States Department ofEnergy, Grant DE-FG02-04ER41286. Part of this work has beenfinanced by the Spanish Ministry of Economy and Competitivenessthrough the Particle Physics National Program (FPA2013-48308-C2-2-P and FPA2013-48387-C6-2-P), by the European Union'sHorizon 2020 Research and Innovation funding program, under

Grant Agreement no. 654168 (AIDA-2020), and by the ItalianMinistero degli Affari Esteri and INFN Gruppo V.

References

[1] H. F.-W. Sadrozinski, Exploring charge multiplication for fast timing with si-licon sensors, 20th RD50 Workshop, Bari, Italy, June 2-6, 2012; ⟨https://indico.cern.ch/event/175330/session/8/contribution/18/attachments/225144/315064/RD50_Bari_UFSD_Sadrozinski.pdf⟩.

[2] G. Pellegrini, et al., Nucl. Instrum. Methods A765 (2014) 12.[3] G. Pellegrini, et al., Recent Technological Developments on LGAD and iLGAD

Detectors for Tracking and Timing Applications these Proceedings.[4] H.F.-W. Sadrozinski, et al., Nucl. Instrum. Methods A765 (2014) 7.[5] F. Cenna, et al., Nucl. Instrum. Methods A796 (2015) 149 ⟨http://personalpages.

to.infn.it/�cartigli/Weightfield2/Main.html⟩.[6] B. Baldassari, et al., Signal formation in irradiated silicon detectors, 14th

Vienna Conference on Instrumentation, Vienna, Austria, Feb. 15–19, 2016.[7] RD50 collaboration, ⟨http://rd50.web.cern.ch/rd50/⟩.[8] G. Kramberger, et al., Radiation effects in Low Gain Avalanche Detectors after

hadron irradiations, JINST 10 (2015) P07006.[9] G. Kramberger, et al., Effects of irradiation on LGAD devices with high excess

current, 25th RD50 Workshop, CERN, Switzerland, Nov 19–21, 2014; ⟨https://indico.cern.ch/event/334251/session/2/contribution/31/attachments/652609/897364/Effect_of_excess_current_to_detector_operation.pdf25th⟩ RD50Workshop, CERN, Switzerland, Nov 11–13, 2014.

[10] N. Cartiglia, et al., Nucl. Instrum. Methods A796 (2015) 141–148.[11] A. Seiden, Ultra-Fast Silicon Detectors, PoS(VERTEX2015) 025, 2015.